Geometric deformable model driven by CoCRFs: application to optical coherence tomography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Geographic atrophy (GA) is a leading cause of vision loss in age-related macular degeneration (AMD).
• Accurate segmentation of GA in Optical Coherence Tomography (OCT) images is crucial for diagnosis and monitoring.
• Current segmentation methods may struggle with the complex shapes and ambiguous boundaries of GA.

Purpose of the Study:

• To develop and validate a novel geometric deformable model for segmenting geographic atrophy in OCT fundus images.
• To integrate an active learning-based collaborative Conditional Random Field (CoCRF) for dynamic probability field estimation.
• To improve the accuracy and robustness of GA segmentation in dry AMD.

Main Methods:

• A geometric deformable model using signed distance functions and C1 continuity constraints.
• Inclusion of a shape prior and a connectivity term in the model's internal energy.
• Dynamic estimation of image probability fields using an active learning CoCRF that assesses neighboring site appearance and classification confidence.

Main Results:

• The proposed method successfully segments geographic atrophy in OCT fundus images.
• The active learning CoCRF dynamically refines probability fields for improved accuracy.
• The model demonstrates robustness in handling feature ambiguities and ensuring shape connectivity.

Conclusions:

• The presented geometric deformable model with dynamic probability field updates offers an effective approach for GA segmentation.
• The active learning strategy within the CoCRF enhances segmentation performance in challenging OCT images.
• This method holds promise for clinical applications in diagnosing and managing dry age-related macular degeneration.